时空上下文抗遮挡视觉跟踪

发布时间：2018-10-16 18:20

【摘要】：目的传统的视觉跟踪方法只考虑了目标本身的特征信息提取,忽略了目标周围稠密的上下文信息。一旦目标本身特征信息提取存在困难,很容易导致跟踪失败。为解决上述问题,提出一种时空上下文抗遮挡视觉跟踪算法(STC-PF)。方法首先,利用目标与局部区域的时空关系学习时空上下文模型;然后,通过上下文先验模型和学习到的时空上下文模型计算置信图;最后,对时空上下文区域进行分块遮挡判别,若遮挡概率小于设定阈值,计算置信图所得最大概率位置即为目标位置;若遮挡概率大于设定阈值,则目标发生遮挡,通过子块匹配和粒子滤波估计目标位置以及运动轨迹,实现不同程度的抗遮挡跟踪。结果对测试数据集中的图像序列进行实验,结果表明,STC-PF方法的跟踪成功率提高至80%以上;中心误差小于原算法;同时STC-PF算法在提高抗遮挡能力的前提下,运行速度与原算法相当,高于当前流行算法。结论 STC-PF算法能够适用于光照变化、目标旋转、遮挡等复杂情况下的视觉目标跟踪,具有一定的实时性和高效性,尤其是在目标发生遮挡情况下具有很好的抗遮挡能力和较快的运行速度。
[Abstract]:Objective the traditional visual tracking method only considers the feature extraction of the target itself and ignores the dense contextual information around the target. Once the feature information extraction of the target itself is difficult, it is easy to lead to the tracking failure. In order to solve the above problems, a spatio-temporal context-resistant visual tracking algorithm (STC-PF) is proposed. Methods first, the spatio-temporal context model is studied by using the spatio-temporal relationship between the target and the local region; then, the confidence diagram is calculated by the contextual priori model and the spatio-temporal context model. If the occlusion probability is less than the set threshold, the maximum probability position of the confidence chart is the target position, and if the occlusion probability is greater than the set threshold, the occlusion occurs. The target position and motion trajectory are estimated by subblock matching and particle filter to achieve different degree of anti-occlusion tracking. Results the experimental results show that the tracking success rate of STC-PF method is more than 80%, the center error is smaller than that of the original algorithm, and the STC-PF algorithm can improve the ability of resisting occlusion. The running speed is similar to the original algorithm, which is higher than the current popular algorithm. Conclusion the STC-PF algorithm can be used to track the visual targets in complex cases, such as illumination change, target rotation, occlusion and so on, and has a certain real-time and high efficiency. Especially in the case of object occlusion, it has good ability of resisting occlusion and fast running speed.
【作者单位】：辽宁工程技术大学软件学院;
【基金】：国家自然科学基金项目(61172144) 辽宁省教育厅科学技术研究一般项目(L2015216)~~
【分类号】：TP391.41

【相似文献】